Z-Image Spicy · Text To Image

From $0.01235 an image, uncensored, no prompt length limit, and - the part that actually separates it from the other Z-Image nodes in this pack - you type the width and height yourself instead of picking a ratio from a dropdown. Width runs 256 to 1536, but height defaults to 1536, so the node is clearly built for tall and poster-shaped work rather than the usual 1:1.

Where it fits

Z-Image earned its place in the local ecosystem by being small, fast and Apache 2.0 at a moment when everything else was getting bigger and more restricted. Hosted, it's the cheap end of image generation: a hundred images for a bit over a dollar. That makes it a fine choice for prompt sweeps, thumbnail batches, idea boards, and compositions where a non-standard frame matters - a 1024×1536 portrait, a 1536×640 banner. If you need maximum quality rather than maximum volume, the Spicy Pro endpoint is the tier above; if you want the standard ratios and roughly a second a frame, the plain Z-Image node is the faster lane.

The inputs

• prompt - straightforward description; subject, composition, style, lighting. No character cap, so write as much as you like.
• width - integer, default 1024, between 256 and 1536.
• height - integer, default 1536, in the same range.
• enable_prompt_expansion - off by default. It rewrites your prompt into a fuller description before generating, which helps lazy prompts and muddies deliberate ones.
• seed - the same seed with the same settings returns the same image, which is what makes A/B testing prompts possible at all.

Outputs: images (IMAGE, so Save Image or anything downstream) and task id (STRING - the SpicyAPI task id, for the console log or support).

The pixel-control caveat

Z-Image's native output sits around 1–2 megapixels, and 1536×1536 is 2.36MP. You can ask for it, but if the frame comes back looking softer than a 1024-wide render, that's not you doing anything wrong. Stay near the native band when detail matters and let a local upscaler do the lifting - it's cheaper and you keep control of the method.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/spicyapi-ai/comfyui-spicyapi

ComfyUI Manager → search SpicyAPI is the other route. Restart ComfyUI, then Settings > SpicyAPI → paste a key from spicyapi.ai/console/keys, and restart once more so the node list is rebuilt from your account. Servers: set SPICY_API_KEY. Nothing else installs - no model downloads and an empty dependency list, so only ComfyUI itself, Python 3.10+ and the V3 node API (tested with ComfyUI 0.39).

Money, seeds and errors

Set Max cost per run in the SpicyAPI settings. Even at a penny an image, the pack's pre-flight quote is the reason a mis-typed queue of two hundred batching runs doesn't just happen. Seeds default to fixed, so queueing the same graph twice gives you the first result again instead of a second charge; flip control after generate to randomize when you want variety, and account for it. And if you cancel a run, the job continues on SpicyAPI and is charged if it succeeds.

Error messages here are specific enough to be useful: No SpicyAPI key is set (add the key, or the env var), 40310 (click the verification link from your signup email), 40004 (no deployment serves that width/height combination - change what the message names), or an error tail with request_id ..., which is exactly what support needs.